Synesthetic Room Impulse Responses Dataset

This Zenodo record provides the Synthetic Room Impulse Responses (RIRs) dataset used to train and evaluate the LSTM-based neural network for EDCs prediction.

The dataset consists of 17,640 RIR files stored in .wav format. Each file corresponds to a unique simulated shoebox room configuration, covering a wide range of room geometries, source–receiver placements, and surface absorption properties.

The RIRs were synthesized using pyroomacoustics (ISM and ray tracing) standard acoustic simulation methods. This dataset used to compute EDCs (https://zenodo.org/records/17210197), which are then used in LSTM model (https://zenodo.org/records/17215057) to training and benchmarking of deep learning models for tasks such as:

• Room acoustic parameter estimation
• EDC prediction from room features
• RIR reconstruction using random sticky-sign technique (https://arxiv.org/abs/2509.24834)

The dataset pairs naturally with the trained LSTM model and inference pipeline available on GitHub:
👉 https://github.com/TUIlmenauAMS/LSTM-Model-Energy-Decay-Curves

Dataset Contents

• Files: 17,640 .wav files
• Sampling rate: 48 kHz
• RIR length: 3 seconds
• File naming: Each filename corresponds to a unique room ID, which maps to the room feature CSV in the repository.

This dataset is intended for research and educational use. Please cite appropriately when used in publications.